PANCS: Proactive self-adaptation for nonlinear cyber–physical systems

Cyber–Physical Systems (CPS) often exhibit nonlinear behavior and are subject to considerable uncertainty arising from dynamic environmental conditions, hardware degradation, and incomplete system knowledge. While reactive self-adaptation frequently fails to maintain functionality and quality under these conditions, proactive adaptation proves effective, leveraging predictive insights. However, most existing proactive approaches are tailored to information systems and lack systematic architectural support for nonlinear CPS with real-time constraints. This paper presents PANCS, a reusable reference architecture grounded in the MAPE–K loop and enhanced with Adaptive Model Predictive Control. PANCS enables proactive runtime adaptation in complex, nonlinear CPS by anticipating system behavior and environmental dynamics. It further integrates learning mechanisms to mitigate model-to-reality discrepancies and ensures real-time feasibility through dynamic horizon adjustment and constraint-aware optimization. PANCS is evaluated through a warehouse robotic case study using distributed co-simulation under realistic uncertainty scenarios. As a reference architecture, PANCS is not domain-specific and its reusability is illustrated in two dimensions; integration within larger adaptive ecosystems through established MAPE–K decentralized patterns, and conceptual instantiation of its architectural components across two additional CPS domains.